Today’s biggest AI thread is a blunt reality check on capacity and returns. At a financial event, Intel CEO Lip-Bu Tan and Broadcom CEO Hock Tan warned that capital access and “AI ROI” concerns may choke some appetite just as compute demand continues to outstrip supply. The most concrete datapoint came from Intel: the company said it can meet only 50% of orders for AI-related CPU demand. That mismatch helps explain why the AI buildout has felt both expensive and constrained—threads of optimism run through the industry, but the supply chain is still doing the math.
The executives didn’t claim relief is immediate. They expect compute shortages to ease by 2028 or 2029, yet also cautioned that the demand–supply imbalance could linger into 2030. Broadcom’s angle adds another constraint beyond hardware: when funding tightens, “AI ROI” becomes an investment committee question, not just a technical one. The practical takeaway is that planning horizons may need to stretch. Even as performance improves, organizations are likely to feel procurement friction and delayed deployments, turning near-term AI strategies into scheduling exercises as much as model-building ones.
